这次发布貌似比较低调，但是发布后，社区反响还是不错，排名前 列，还在持续上升中，我们也聊聊这次的亮点 核心性能突破 超长上下文处理 技术实现： 128K tokens的工程突破与架构优化 DeepSeek V3.1将上下文窗口扩展至128K tokens，实现对上一代版本（64K）的翻倍提升，可处理约10 万-13万汉字（相当于两本200页小说或400页书籍的文本量）。这一突破的核心在于 Transformer架构 的内存管理优化 ，通过改进注意力机制的计算效率与上下文状态追踪能力，解决了前代模型在长文本 处理中常见的 上下文丢失 与 响应碎片化 问题。线上模型版本与开源版本保持一致的上下文能力，确保 企业级用户与开发者可获得同等的长文本处理性能。 场景验证：从长文档分析到复杂任务支持 在企业级应用中， 128K上下文能力显著提升了 法律合同审查 、 学术论文综述 等场景的效率。模型可 一次性输入完整的超长法律文档（如 400页合同）或博士论文（约10万汉字），并保持逻辑连贯性与细 节准确性。实测显示，其在约10万字 文章中 删减文本中成功定位到特定句子，并 能理解文章内容 ，验 证了长文本中的 精准信息检索 ...

Financial Data and Key Metrics Changes - The company reported its first profitable second quarter with operating income of RMB28.8 million compared to an operating loss of RMB72.6 million in the same period last year [6] - Net revenues reached RMB1.4 billion, an increase of 7.2% year over year [6][20] - Operating cash inflow was RMB185 million, down 26.1% year over year, primarily due to strategic scaling back of certain courses [7] - Total gross profit was RMB609.4 million, representing a 4.3% decrease from the same period of 2024 [21] - Non-GAAP net income attributable to ordinary shareholders was RMB12.5 million compared to a non-GAAP net loss of RMB96 million for the same period last year [23] Business Line Data and Key Metrics Changes - Net revenues from learning services rose 2.2% year over year to RMB657.8 million, driven by strong performance in Youdao Ling Shi [7][21] - Net revenues from online marketing services reached RMB632.9 million, up 23.8% year over year, driven by demand from the gaming industry and overseas markets [12][21] - Net revenues from smart devices declined 23.9% year over year to RMB126.8 million, attributed to the end of product life cycles and reduced marketing expenditure [15][21] Market Data and Key Metrics Changes - The gaming advertising segment saw revenue growth of more than 50% year over year, supported by collaborations with major gaming advertisers [13] - The overseas market contributed significantly to growth, with revenue from partnerships with TikTok and Google increasing significantly [64] Company Strategy and Development Direction - The company aims to advance its AI native strategy, focusing on scenario-based optimizations of large language models to enhance learning and advertising services [18] - There is a strong emphasis on integrating hardware and learning services to improve operational efficiency and reduce sales and marketing expenses [40] Management's Comments on Operating Environment and Future Outlook - Management expressed confidence in achieving operating cash flow breakeven despite a year-over-year decline in operating cash inflow [52][56] - The company anticipates stronger cash flow performance in the second half of the year, driven by improved profitability and operational efficiency [54] Other Important Information - The company launched several AI-driven features and products, including the AI essay grading feature and the Confucius III language model, which received positive feedback [8][10] - The company signed 12 gold medalists from the National Olympiads in Informatics to enhance its teaching and R&D capabilities [9] Q&A Session Summary Question: Update on the third quarter outlook for Youdao Ling Shi - Management noted that Youdao Ling Shi's revenue increased by roughly 30% year over year, with a retention rate exceeding 75%, indicating strong user satisfaction and a solid foundation for future growth [28][30] Question: Improvement in Smart Device segment revenue - Management stated that while revenue declined in Q2, the health of the hardware business improved compared to the previous year, with a focus on dictionary pens and new tutoring pens expected to drive future growth [36][39] Question: Specific applications of AI ad placement optimizer - The AI ad placement optimizer covers the entire advertising delivery process, enhancing targeting strategies and optimizing ad delivery, which is expected to support revenue growth and profitability improvement [44][48] Question: Revision on the target for achieving operating cash flow breakeven - Management confirmed that despite a decrease in operating cash flow, the target for achieving breakeven remains unchanged, supported by improved profitability and operational efficiency [52][56] Question: Growth drivers in gaming and overseas markets - Management highlighted a 50% year-over-year increase in gaming revenue and significant growth in overseas markets, particularly through partnerships with TikTok and Google [63][64]

A Malaysian company has designed an AI large language model for Muslims based on open-source AI knowhow from China’s DeepSeek https://t.co/in9zO3EGy7 ...

[Music] Welcome to the Vergecast, the flagship podcast of testing cursed technology. I'm your friend V Song and I'm here with a special Sunday bonus episode. Yay.We're calling this AI for normies. So, here's the concept. AI can be so open-ended, it's really hard for the average person to know what it's good for.And if you ask me, I don't think big tech is doing such a great job at explaining that either. But we here at the verge. com are a bunch of giant nerds and we test all of this stuff for a living.So I ...

We’re proud to present zkGPT, a system for proving that large language model inference was performed correctly, without revealing the model.It enables private, verifiable LLM inference and generates compact proofs in under 25 seconds.The paper is now live: https://t.co/F32TZ9UewR ...

Core Insights - Innodata Inc. (INOD) is positioned as a key player in the AI revolution by providing essential data for training advanced language models [1] - The company reported Q2 2025 adjusted earnings per share of $0.20, exceeding the Zacks Consensus Estimate of $0.11 [1] - Quarterly revenues reached $58.39 million, reflecting a 79% year-over-year increase and surpassing estimates by 3.6% [2] Revenue Growth and Guidance - Following strong Q2 performance, Innodata raised its 2025 revenue growth guidance to over 45% year-over-year, up from a previous forecast of 40% [2] - The expected revenue growth rate for the current year is 41.9%, while the earnings growth rate is projected at -23.6% [6] AI Demand and Market Position - Innodata is set to benefit from the increasing demand for data engineering services in large language model development, supporting five of the seven major hyperscalers [3] - The company has diversified its customer base, which is expected to support long-term growth across various sectors including technology, healthcare, and federal agencies [4] New Product Launch - Innodata introduced a GenAI Test and Evaluation Platform aimed at validating large language models, with MasterClass as the first customer [5] - The platform is designed to enhance integration with major tech companies' upcoming GenAI investments [5] Stock Performance and Estimates - Innodata's stock is currently trading 38.6% below its 52-week high, despite a year-to-date return of 10.3%, outperforming the S&P 500 [7] - Brokerage targets suggest a potential upside of 72.1%, with average short-term price targets indicating a 53.2% increase from the last closing price of $43.58 [10] Consensus Estimates - The Zacks Consensus Estimate for current-year earnings has remained stable over the last 30 days, while next-year earnings estimates have improved by 2.9% [6]

Core Insights - The article discusses advancements in trajectory prediction using a generative active learning framework called GALTraj, which applies controllable diffusion models to address long-tail issues in data [1][2]. Group 1: GALTraj Framework - GALTraj is the first framework to apply generative active learning to trajectory prediction tasks, enhancing long-tail learning without modifying the model structure [2]. - The framework employs a tail-aware generation method that differentiates the diffusion guidance for tail, head, and related agents, producing realistic and diverse scenarios while preserving tail characteristics [2][3]. Group 2: Experimental Results - In experiments on WOMD and Argoverse2 datasets, GALTraj significantly improved long-tail sample prediction performance, reducing the long-tail metric FPR₅ by 47.6% (from 0.42 to 0.22) and overall prediction error minFDE₆ by 14.7% (from 0.654 to 0.558) [1][6]. - The results indicate that GALTraj outperforms traditional methods across various metrics, showcasing its effectiveness in enhancing prediction accuracy for rare scenarios [7][8]. Group 3: TopoLiDM Framework - The article also highlights the TopoLiDM framework developed by Shanghai Jiao Tong University and Twente University, which integrates topology-aware diffusion models for high-fidelity LiDAR point cloud generation [13][15]. - TopoLiDM achieved a 22.6% reduction in the Fréchet Range Image Distance (FRID) and a 9.2% reduction in Minimum Matching Distance (MMD) on the KITTI-360 dataset while maintaining a real-time generation speed of 1.68 samples per second [13][15]. Group 4: FastDriveVLA Framework - FastDriveVLA, developed by Peking University and Xiaopeng Motors, introduces a reconstruction-based visual token pruning framework that maintains 99.1% trajectory accuracy with a 50% pruning rate and reduces collision rates by 2.7% [21][22]. - The framework employs a novel adversarial foreground-background reconstruction strategy to enhance the identification of valuable tokens, achieving state-of-the-art performance on the nuScenes open-loop planning benchmark [27][28]. Group 5: PLA Framework - The article presents a unified Perception-Language-Action (PLA) framework proposed by TUM, which integrates multi-sensor fusion and GPT-4.1 enhanced visual-language-action reasoning for adaptive autonomous driving [34][35]. - The framework demonstrated a mean absolute error (MAE) of 0.39 m/s in speed prediction and an average displacement error (ADE) of 1.013 meters in trajectory tracking within urban intersection scenarios [42].

Group 1: Core Insights - The article emphasizes the importance of foundational skills in programming and software engineering for entering the AI field, with Python being the preferred language due to its ease of use and comprehensive ecosystem [1][2][4] - It highlights that while many AI roles stem from machine learning, the most sought-after positions are closer to software engineering, necessitating knowledge of languages like Java, GO, or Rust [1][2] - Continuous practice and real-world application are deemed essential for mastering programming languages, rather than solely relying on courses or books [2] Group 2: Recommended Resources - A variety of resources are suggested for learning Python, including a beginner's course that can be completed in four hours and a highly regarded specialization course [5] - For mathematics and statistics, specific books and courses are recommended to understand the underlying principles of machine learning and AI [9][10] - The article lists essential resources for deep learning and large language models, emphasizing the significance of frameworks like PyTorch and TensorFlow in the industry [13][14] Group 3: AI Engineering and Productization - The article stresses the need for skills in productizing AI models, indicating that most AI roles resemble traditional software engineering rather than pure machine learning engineering [11] - It mentions the importance of learning MLOps for model deployment, covering aspects like containerization and cloud systems [11] - The article concludes with advice on becoming an expert in the field through project-based learning and self-reflection [14]

机器之心报道 机器之心编辑部 AI 一定会比人类更聪明，之后会发生什么？ 今天上午，在世界人工智能大会 WAIC 上，2024 年诺贝尔物理学奖得主、2018 年图灵奖得主、人工智能教父杰弗里・辛顿（Geoffrey Hinton）发表了题为「 数字 智能是否会取代生物智能 」的开场演讲。 该演讲围绕人工智能领域的历史、未来发展方向、语言模型的原理、数字与生物计算特点以及 AI 发展带来的担忧等内容展开，辛顿高度评价了当前 AI 领域的大 模型技术，认为其与人类思考模式相同。 大语言模型，在用人类的方式思考？ 非常感谢大家给我这样一个机会，让我来分享一下个人的观点 —— 有关 AI 的历史和它的未来。 在过去 60 多年来，学界对于 AI 有两种不同的理解范式，一个是逻辑型，认为符号规则的表达操作可以实现推理；另一种是图灵和冯诺依曼所相信的，认为智能 的基础在于学习神经网络中的链接，这个过程中理解是第一位的。 这让我们开始关注语言中词与词之间的关系。 心理学家有另一套理论，他们认为数字是语义学的特征。在 1985 年，我做了一个很小的模型，想把两大理论方向结合在一起，来更好地理解人类是如何理解词汇 的。我对每 ...

Core Viewpoint - The article highlights a significant achievement in artificial intelligence (AI), where large language models (LLMs) have reached gold medal level in the International Mathematical Olympiad (IMO), showcasing their advanced problem-solving capabilities [4][5][6]. Group 1: AI Achievement - Google DeepMind's large language model successfully solved problems equivalent to those in the IMO, achieving a score that surpasses the gold medal threshold of 35 out of 42 [4][5]. - This marks a substantial leap from the previous year's performance, where the model was only at the silver medal level, indicating a qualitative breakthrough in AI's ability to handle complex mathematical reasoning [5][6]. Group 2: Implications of the Achievement - The success of LLMs in the IMO demonstrates their capability to tackle highly complex tasks that require deep logical thinking and abstract reasoning, beyond mere text generation [7]. - Such AI advancements can serve as powerful tools in education and research, assisting students in learning higher mathematics and aiding researchers in exploring new conjectures and theorems [7]. - Achieving gold medal level in mathematics is a significant milestone on the path to artificial general intelligence (AGI), as it requires a combination of various cognitive abilities [7][8]. Group 3: Broader Impact - The breakthroughs by DeepMind and OpenAI not only elevate AI's status in mathematical reasoning but also suggest vast potential for future applications in scientific exploration and technological development [8].